Forecasting summertime surface-level ozone concentrations in the Lower Fraser Valley of British Columbia: an ensemble neural network approach

Empirical models for predicting daily maximum hourly average ozone concentrations were developed for 10 monitoring stations in the Lower Fraser Valley (LFV) of British Columbia. According to data from 1991 to 1996, ensemble neural network models increased explained variance an average of 7% over multiple linear regression models using the same input variables. Without modification, all models performed poorly on days when the observed peak ozone concentration exceeded 82 parts per billion, the National Ambient Air Quality Objective. When numbers of extreme events in training data were increased using a histogram equalization process, models were able to forecast exceedances with improved accuracy. Modified generalized additive model (GAM) plots and associated measures of input variable importance and interaction were generated for a subset of the trained models and used to investigate relationships between input variables and ozone levels. The neural network models displayed a high degree of interaction among inputs, and it is likely the ability of these model types to account for interactions, rather than the nonlinearity of individual input variables, that explains their improved forecast skill. Inspection of GAM-style plots indicated that the relative importance of input variables in the ensemble neural network models varied with geographic location within the LFV. Four distinct groups of stations were identified, and rankings of inputs within the groups were generally consistent with physical intuition and results of prior studies.     

Multi-model vs. EPS-based ensemble atmospheric dispersion simulations: A quantitative assessment on the ETEX-1 tracer experiment case

Several techniques have been developed over the last decade for the ensemble treatment of atmospheric dispersion model predictions. Among them two have received most of the attention, the multi-model and the ensemble prediction system (EPS) modeling. The multi-model approach relies on model simulations produced by different atmospheric dispersion models using meteorological data from potentially different weather prediction systems. The EPS-based ensemble is generated by running a single atmospheric dispersion model with the ensemble weather prediction members. In the paper we compare both approaches with the help of statistical indicators, using the simulations performed for the ETEX-1 tracer experiment. Both ensembles are also evaluated against measurement data. Among the most relevant results is that the multi-model median and the mean of EPS-based ensemble produced the best results, hence we consider a combination of multi-model and EPS-based approaches as an interesting suggestion for further research.     

Skill and uncertainty of a regional air quality model ensemble

Recently several regional air quality projects were carried out to support the negotiation under the Clean Air For Europe (CAFE) programme by predicting the impact of emission control policies with an ensemble of models. Within these projects, CITYDELTA and EURODELTA, the fate of air quality at the scale of European cities or that of the European continent was studied using several models. In this article we focus on the results of EURODELTA. The predictive skill of the ensemble of models is described for ozone, nitrogen dioxide and secondary inorganic compounds, and the uncertainty in air quality modelling is examined through the model ensemble spread of concentrations.For ozone daily maxima the ensemble spread origin differs from one region to another. In the neighbourhood of cities or in mountainous areas the spread of predicted values does not span the range of observed data, due to poorly resolved emissions or complex-terrain meteorology. By contrast in Atlantic and North Sea coastal areas the spread of predicted values is found to be larger than the observations. This is attributed to large differences in the boundary conditions used in the different models. For NO₂ daily averages the ensemble spread is generally too small compared with observations. This is because models miss highest values occurring in stagnant meteorology in stable boundary layers near cities. For secondary particulate matter compounds the simulated concentration spread is more balanced, observations falling nearly equiprobably within the ensemble, and the spread originates both from meteorology and aerosol chemistry and thermodynamics.     

A novel bagging ensemble approach for predicting summertime ground-level ozone concentration

Ozone pollution appears as a major air quality issue, e.g. for the protection of human health and vegetation. Formation of ground level ozone is a complex photochemical phenomenon and involves numerous intricate factors most of which are interrelated with each other. Machine learning techniques can be adopted to predict the ground level ozone. The main objective of the present study is to develop the state-of-the-art ensemble bagging approach to model the summer time ground level ozone in an industrial area comprising a hazardous waste management facility. In this study, the feasibility of using ensemble model with seven meteorological parameters as input variables to predict the surface level O3 concentration. Multilayer perceptron, RTree, REPTree, and Random forest were employed as the base learners. The error measures used for checking the performance of each model includes IoAd, R2, and PEP. The model results were validated against an independent test data set. Bagged random forest predicted the ground level ozone better with higher Nash-Sutcliffe coefficient 0.93. This study scaffolded the current research gap in big data analysis identified with air pollutant prediction.

Implications: The main focus of this paper is to model the summer time ground level O₃ concentration in an Industrial area comprising of hazardous waste management facility. Comparison study was made between the base classifiers and the ensemble classifiers. Most of the conventional models can well predict the average concentrations. In this case the peak concentrations are of importance as it has serious effect on human health and environment. The models developed should also be homoscedastic.     

Ozone risk assessment for agricultural crops in Europe: Further development of stomatal flux and flux–response relationships for European wheat and potato

Applications of a parameterised Jarvis-type multiplicative stomatal conductance model with data collated from open-top chamber experiments on field grown wheat and potato were used to derive relationships between relative yield and stomatal ozone uptake. The relationships were based on thirteen experiments from four European countries for wheat and seven experiments from four European countries for potato. The parameterisation of the conductance model was based both on an extensive literature review and primary data. Application of the stomatal conductance models to the open-top chamber experiments resulted in improved linear regressions between relative yield and ozone uptake compared to earlier stomatal conductance models, both for wheat (r²=0.83) and potato (r²=0.76). The improvement was largest for potato. The relationships with the highest correlation were obtained using a stomatal ozone flux threshold. For both wheat and potato the best performing exposure index was AF_st6 (accumulated stomatal flux of ozone above a flux rate threshold of 6 nmol ozone m⁻² projected sunlit leaf area, based on hourly values of ozone flux). The results demonstrate that flux-based models are now sufficiently well calibrated to be used with confidence to predict the effects of ozone on yield loss of major arable crops across Europe. Further studies, using innovations in stomatal conductance modelling and plant exposure experimentation, are needed if these models are to be further improved.     

Particulate air quality model predictions using prognostic vs. diagnostic meteorology in central California

Comparisons were made between three sets of meteorological fields used to support air quality predictions for the California Regional Particulate Air Quality Study (CRPAQS) winter episode from December 15, 2000 to January 6, 2001. The first set of fields was interpolated from observations using an objective analysis method. The second set of fields was generated using the WRF prognostic model without data assimilation. The third set of fields was generated using the WRF prognostic model with the four-dimensional data assimilation (FDDA) technique. The UCD/CIT air quality model was applied with each set of meteorological fields to predict the concentrations of airborne particulate matter and gaseous species in central California. The results show that the WRF model without data assimilation over-predicts surface wind speed by ～30% on average and consequently yields under-predictions for all PM and gaseous species except sulfate (S(VI)) and ozone(O₃). The WRF model with FDDA improves the agreement between predicted and observed wind and temperature values and consequently yields improved predictions for all PM and gaseous species. Overall, diagnostic meteorological fields produced more accurate air quality predictions than either version of the WRF prognostic fields during this episode. Population-weighted average PM_2.5 exposure is 40% higher using diagnostic meteorological fields compared to prognostic meteorological fields created without data assimilation. These results suggest diagnostic meteorological fields based on a dense measurement network are the preferred choice for air quality model studies during stagnant periods in locations with complex topography.     

Model for forecasting expressway fine particulate matter and carbon monoxide concentration: application of regression and neural network models

The Borman Expressway is a heavily traveled 16-mi segment of the Interstate 80/94 freeway through Northwestern Indiana. The Lake and Porter counties through which this expressway passes are designated as particulate matter < 2.5 microm (PM2.5) and ozone 8-hr standard nonattainment areas. The Purdue University air quality group has been collecting PM2.5, carbon monoxide (CO), wind speed, wind direction, pressure, and temperature data since September 1999. In this work, regression and neural network models were developed for forecasting hourly PM2.5 and CO concentrations. Time series of PM2.5 and CO concentrations, traffic data, and meteorological parameters were used for developing the neural network and regression models. The models were compared using a number of statistical quality indicators. Both models had reasonable accuracy in predicting hourly PM2.5 concentration with coefficient of determination -0.80, root mean square error (RMSE) <4 microg/m3, and index of agreement (IA) > 0.90. For CO prediction, both models showed moderate forecasting performance with a coefficient of determination -0.55, RMSE < 0.50 ppm, and IA -0.85. These models are computationally less cumbersome and require less number of predictors as compared with the deterministic models. The availability of real time PM2.5 and CO forecasts will help highway managers to identify air pollution episodic events beforehand and to determine mitigation strategies.     

Evaluation of artificial neural networks for fine particulate pollution (PM10 and PM2.5) forecasting

Multi-layer perceptron (MLP) artificial neural network (ANN) models are compared with traditional multiple regression (MLR) models for daily maximum and average O3 and particulate matter (PM10 and PM2.5) forecasting. MLP particulate forecasting models show little if any improvement over MLR models and exhibit less skill than do O3 forecasting models. Meteorological variables (precipitation, wind, and temperature), persistence, and co-pollutant data are shown to be useful PM predictors. If MLP approaches are adopted for PM forecasting, training methods that improve extreme value prediction are recommended.     

CityDelta: A model intercomparison study to explore the impact of emission reductions in European cities in 2010

This paper gives an overview of the set up, methodology and the obtained results of the CityDelta (phase 1 and 2) project. In the context of the Clean Air For Europe programme of the European Commission, the CityDelta project was designed to evaluate the impact of emission-reduction strategies on air quality at the European continental scale and in European cities. Ozone and particulate matter (PM) are the main components that have been studied. To achieve this goal, a model intercomparison study was organized with the participation of more than 20 modelling groups with a large number of modelling configurations. Two following main topics can be identified in the project. First, in order to evaluate their strengths and weaknesses, the participating models were evaluated against observations in a control year (1999). An accompanying paper will discuss in detail this evaluation aspect for four European cities. The second topic is the actual evaluation of the impact of emission reductions on levels of ozone and PM, with particular attention to the differences between large-scale and fine-scale models. An accompanying paper will discuss this point in detail. In this overview paper the main input to the intercomparison is described as well as the use of the ensemble approach. Finally, attention is given to the policy relevant issue on how to implement the urban air quality signal into large-scale air quality models through the use of functional relationships.     

Parameter evaluation and model validation of ozone exposure assessment using Harvard Southern California Chronic Ozone Exposure Study data

To examine factors influencing long-term ozone (O3) exposures by children living in urban communities, the authors analyzed longitudinal data on personal, indoor, and outdoor O3 concentrations, as well as related housing and other questionnaire information collected in the one-year-long Harvard Southern California Chronic Ozone Exposure Study. Of 224 children contained in the original data set, 160 children were found to have longitudinal measurements of O3 concentrations in at least six months of 12 months of the study period. Data for these children were randomly split into two equal sets: one for model development and the other for model validation. Mixed models with various variance-covariance structures were developed to evaluate statistically important predictors for chronic personal ozone exposures. Model predictions were then validated against the field measurements using an empirical best-linear unbiased prediction technique. The results of model fitting showed that the most important predictors for personal ozone exposure include indoor O3 concentration, central ambient O3 concentration, outdoor O3 concentration, season, gender, outdoor time, house fan usage, and the presence of a gas range in the house. Hierarchical models of personal O3 concentrations indicate the following levels of explanatory power for each of the predictive models: indoor and outdoor O3 concentrations plus questionnaire variables, central and indoor O3 concentrations plus questionnaire variables, indoor O3 concentrations plus questionnaire variables, central O3 concentrations plus questionnaire variables, and questionnaire data alone on time activity and housing characteristics. These results provide important information on key predictors of chronic human exposures to ambient O3 for children and offer insights into how to reliably and cost-effectively predict personal O3 exposures in the future. Furthermore, the techniques and findings derived from this study also have strong implications for selecting the most reliable and cost-effective exposure study design and modeling approaches for other ambient pollutants, such as fine particulate matter and selected urban air toxics.     

Impact of biogenic emission uncertainties on the simulated response of ozone and fine particulate matter to anthropogenic emission reductions

The role of emissions of volatile organic compounds and nitric oxide from biogenic sources is becoming increasingly important in regulatory air quality modeling as levels of anthropogenic emissions continue to decrease and stricter health-based air quality standards are being adopted. However, considerable uncertainties still exist in the current estimation methodologies for biogenic emissions. The impact of these uncertainties on ozone and fine particulate matter (PM2.5) levels for the eastern United States was studied, focusing on biogenic emissions estimates from two commonly used biogenic emission models, the Model of Emissions of Gases and Aerosols from Nature (MEGAN) and the Biogenic Emissions Inventory System (BEIS). Photochemical grid modeling simulations were performed for two scenarios: one reflecting present day conditions and the other reflecting a hypothetical future year with reductions in emissions of anthropogenic oxides of nitrogen (NOx). For ozone, the use of MEGAN emissions resulted in a higher ozone response to hypothetical anthropogenic NOx emission reductions compared with BEIS. Applying the current U.S. Environmental Protection Agency guidance on regulatory air quality modeling in conjunction with typical maximum ozone concentrations, the differences in estimated future year ozone design values (DVF) stemming from differences in biogenic emissions estimates were on the order of 4 parts per billion (ppb), corresponding to approximately 5% of the daily maximum 8-hr ozone National Ambient Air Quality Standard (NAAQS) of 75 ppb. For PM2.5, the differences were 0.1-0.25 microg/m3 in the summer total organic mass component of DVFs, corresponding to approximately 1-2% of the value of the annual PM2.5 NAAQS of 15 microg/m3. Spatial variations in the ozone and PM2.5 differences also reveal that the impacts of different biogenic emission estimates on ozone and PM2.5 levels are dependent on ambient levels of anthropogenic emissions.     

The MICS-Asia study: model intercomparison of long-range transport and sulfur deposition in East Asia

An intercomparison study involving eight long-range transport models for sulfur deposition in East Asia has been initiated. The participating models included Eulerian and Lagrangian frameworks, with a wide variety of vertical resolutions and numerical approaches. Results from this study, in which models used common data sets for emissions, meteorology, and dry, wet and chemical conversion rates, are reported and discussed. Model results for sulfur dioxide and sulfate concentrations, wet deposition amounts, for the period January and May 1993, are compared with observed quantities at 18 surface sites in East Asia. At many sites the ensemble of models is found to have high skill in predicting observed quantities. At other sites all models show poor predictive capabilities. Source–receptor relationships estimated by the models are also compared. The models show a high degree of consistency in identifying the main source–receptor relationships, as well as in the relative contributions of wet/dry pathways for removal. But at some locations estimated deposition amounts can vary by a factor or 5. The influence of model structure and parameters on model performance is discussed. The main factors determining the deposition fields are the emissions and underlying meteorological fields. Model structure in terms of vertical resolution is found to be more important than the parameterizations used for chemical conversion and removal, as these processes are highly coupled and often work in compensating directions.     

A study of the association between daily mortality and ambient air pollutant concentrations in Pittsburgh, Pennsylvania

We have studied the possible association of daily mortality with ambient pollutant concentrations (PM10, CO, O3, SO2, NO2, and fine [PM2.5] and coarse PM) and weather variables (temperature and dew point) in the Pittsburgh, PA, area for two age groups--less than 75, and 75 and over--for the 3-year period of 1989-1991. Correlation functions among pollutant concentrations show important seasonal dependence, and this fact necessitates the use of seasonal models to better identify the link between ambient pollutant concentrations and daily mortality. An analysis of the seasonal model results for the younger-age group reveals significant multicollinearity problems among the highly correlated concentrations of PM10, CO, and NO2 (and O3 in spring and summer), and calls into question the rather consistent results of the single- and multi-pollutant non-seasonal models that show a significant positive association between PM10 and daily mortality. For the older-age group, dew point consistently shows a significant association with daily mortality in all models. Collinearity problems appear in the multi-pollutant seasonal and non-seasonal models such that a significant, positive PM10 coefficient is accompanied by a significant, negative coefficient of another ambient pollutant, and the identity of this other pollutant changes with season. The PM2.5 data set is half that of PM10. Identical-model runs for both data sets reveal instability in the pollutant coefficients, especially for the younger age group. The concern for the instability of the pollutant coefficients due to a small signal-to-noise ratio makes it impossible to ascertain credibly the relative associations of the fine- and coarse-particle modes with daily mortality. In this connection, we call for caution in the interpretation of model results for causal inference when the models use fully or partially estimated PM values to fill large data gaps.     

Combining neural network models to predict spatial patterns of airborne pollutant accumulation in soils around an industrial point emission source

Neural networks (NNs) have the ability to model a wide range of complex nonlinearities. A major disadvantage of NNs, however, is their instability, especially under conditions of sparse, noisy, and limited data sets. In this paper, different combining network methods are used to benefit from the existence of local minima and from the instabilities of NNs. A nonlinear k-fold cross-validation method is used to test the performance of the various networks and also to develop and select a set of networks that exhibits a low correlation of errors. The various NN models are applied to estimate the spatial patterns of atmospherically transported and deposited lead (Pb) in soils around an historical industrial air emission point source. It is shown that the resulting ensemble networks consistently give superior predictions compared with the individual networks because, for the ensemble networks, R2 values were found to be higher than 0.9 while, for the contributing individual networks, values for R2 ranged between 0.35 and 0.85. It is concluded that combining networks can be adopted as an important component in the application of artificial NN techniques in applied air quality studies.     

A statistical model for FEV1 response to arbitrary dynamic ozone exposure conditions

Lung function response to inhaled ozone at ambient air pollution levels is known to be a function of ozone concentration, exposure duration, and minute ventilation. Most data-driven exposure-response models address exposures under static condition (i.e., with a constant ozone concentration and exercise pattern). Such models are simplifications, as both ambient ozone concentrations and normal human activity patterns change with time. The purpose of this study was to develop a dynamic model of response with the advantages of a statistical model (a relatively simple structure with few parameters). A previously proposed mechanistic model for changes in specific airways resistance was adapted to describe the percent change in forced expiratory volume in one second (FEV1). This model was then reduced using the fit to three existing exposure-response data sets as criterion. The resulting model consists of a single linear differential equation together with an algebraic logistic equation. Under restricted static conditions the model reduces to a logistic model presented earlier by the authors.     

Extreme value modeling for the analysis and prediction of time series of extreme tropospheric ozone levels: a case study

The occurrence of high concentrations of tropospheric ozone is considered as one of the most important issues of air management programs. The prediction of dangerous ozone levels for the public health and the environment, along with the assessment of air quality control programs aimed at reducing their severity, is of considerable interest to the scientific community and to policy makers. The chemical mechanisms of tropospheric ozone formation are complex, and highly variable meteorological conditions contribute additionally to difficulties in accurate study and prediction of high levels of ozone. Statistical methods offer an effective approach to understand the problem and eventually improve the ability to predict maximum levels of ozone. In this paper an extreme value model is developed to study data sets that consist of periodically collected maxima of tropospheric ozone concentrations and meteorological variables. The methods are applied to daily tropospheric ozone maxima in Guadalajara City, Mexico, for the period January 1997 to December 2006. The model adjusts the daily rate of change in ozone for concurrent impacts of seasonality and present and past meteorological conditions, which include surface temperature, wind speed, wind direction, relative humidity, and ozone. The results indicate that trend, annual effects, and key meteorological variables along with some interactions explain the variation in daily ozone maxima. Prediction performance assessments yield reasonably good results.     

Factors influencing the ground level distribution of ozone in Europe

Ozone is a widely distributed pollutant in the atmospheric boundary layer over north west Europe. Three main sources have been identified: the stratosphere, the free troposphere and boundary layer photochemical production. The pattern of ground level ozone concentrations resulting from these three sources cannot be accurately specified. Ozone shows significant variations in space and time but because of the high cost of continuous monitoring equipment, spatial variations on a national and international basis have not been studied in detail. Variations in ozone concentrations at individual monitoring sites have been given a great deal of attention and experience gained from United Kingdom monitoring sites is described in some detail. The averaging time statistical model of Larsen is employed to relate the exposure levels measured over different averaging periods. Diurnal variations have a major influence on exposure levels at sites nominally exposed to the same regional ozone distribution. The physical and chemical mechanisms which give rise to diurnal variations are detailed so that sites can be screened for different diurnal behaviour characteristics.